math-skill 3.1.0 → 3.2.0

package/skills/math-research-activator/SKILL.en.md

@@ -1,7 +1,7 @@
 ---
 name: math-research-activator
 description: |
-  Mathematical research OS — auto-diagnoses user intent, routes to thinking lenses, math knowledge base, or design translation layer. Triggers on architecture/operator design, theoretical analysis, math-to-AI transfer. Does NOT trigger for pure engineering tasks (debug, refactoring, hyperparameter tuning).
+  Mathematical research OS — auto-diagnoses user intent, routes to thinking lenses, activation anchors, or design translation layer. Triggers on architecture/operator design, theoretical analysis, math-to-AI transfer, and cryptographic definitions, constructions, reductions, or protocol analysis. Does NOT trigger for pure engineering tasks (debug, refactoring, hyperparameter tuning).
 ---
 
 
@@ -11,7 +11,7 @@ description: |
 
 > "The thinking system does not hand out theorems, the knowledge system does not indulge in loose inspiration, and the design layer does not fake profundity."
 
-This system is a mathematical staff office for AI architecture innovation — not an arsenal, but one that tells you: **what kind of battle this is, which arms to deploy, how to deploy them, and where things could go wrong.**
+This system is a mathematical staff office for AI architecture innovation and cryptographic research — not an arsenal, but one that tells you: **what kind of battle this is, which arms to deploy, how to deploy them, and where things could go wrong.**
 
 ## Core Principle
 
@@ -26,36 +26,15 @@ This system is a mathematical staff office for AI architecture innovation — no
 | Layer | Responsibility | Directory | Core Question |
 |-------|---------------|-----------|--------------|
 | **Thinking Lenses** | Diagnose problem structure, recommend mathematical perspectives | `../../lenses/*.en.md` | Which perspective should we view this problem through? |
-| **Math Knowledge** | Provide concrete mathematical tools (definitions/theorems/formulas) | `../../knowledge-base/*/*.en.md` | What specific mathematics does this perspective require? |
+| **Activation Anchors** | Activate high-frequency math structures; trigger Knowledge Gap Protocol when insufficient | `../../knowledge-base/*/*.en.md` | What math structures does this perspective require? |
 | **Design Translation** | Translate mathematics into AI modules/losses/operators | `../../design-patterns/*/*.en.md` | How does this mathematics become model architecture? |
 
 Auxiliary layers:
 - `../../references/books/*.en.md`: Distilled notes from 7 textbooks; full context when deeper understanding is needed
+- `../../references/books/applied-cryptography.md`, `../../references/books/foundations-of-cryptography.md`, `../../references/books/introduction-to-modern-cryptography.md`: 3 English-language cryptography distillations; see `../../references/skill-index.en.md`
 - `../../references/gpu-friendly-math.en.md`: GPU Eight-Dimension Acceptance Gate (single source of truth)
 - `../../agents/math-critic.en.md`: Math-engineering dual critic
 
-## Intent Diagnosis (5 Scenarios)
-
-| Scenario | Diagnostic Signal | Invocation Path |
-|----------|------------------|-----------------|
-| **A. Problem Analysis** | "Is this design sound?" "Are there gaps in the reasoning chain?" | Lenses → critic |
-| **B. Mechanism Design** | "Design a new attention mechanism" "Transfer X to Y" | Lenses → Knowledge → Design → critic |
-| **C. Knowledge Query** | "What is a tangent space on a manifold?" "How is the projection theorem applied?" | Knowledge |
-| **D. Verification & Review** | "Does this formula hold?" "What guarantees does this loss provide?" | Knowledge → relevant design patterns (if a concrete AI construct is cited) → critic |
-| **E. Pure Engineering** | Debugging, refactoring, hyperparameter tuning, code review | **Do not invoke the math system** |
-
-## Lens Library (15 Mathematical Perspectives)
-
-15 lenses covering: Axiomatization, Duality, Symmetry, Spectral Decomposition, Geometric, Projection & Decomposition, Variational, Local-to-Global, Topological, Categorical, Perturbation, Causal, Game-Theoretic, Probabilistic & Statistical, Algorithmic. Directory: `../../lenses/*.en.md`. Full catalog table: `../../references/skill-index.en.md`.
-
-## Knowledge Base (Organized by Mathematical Domain)
-
-7 domains: Matrix Analysis, Optimization, Differential Geometry, Lie Theory, Topology, Probability & Information, Information Geometry. Directory: `../../knowledge-base/*/*.en.md`. Full catalog table: `../../references/skill-index.en.md`.
-
-## Design Pattern Library (Organized by AI Component)
-
-5 component types: Attention, Loss Functions, Routing, Representation, Compression. Directory: `../../design-patterns/*/*.en.md`. Full catalog table: `../../references/skill-index.en.md`.
-
 ## Automatic Trigger Conditions
 
 **All of Gate 1 + Gate 2 + Gate 3 must be satisfied simultaneously for intervention:**
@@ -64,36 +43,85 @@ Auxiliary layers:
 The following tasks **never** trigger the system regardless of workspace contents: code review, debugging, refactoring, hyperparameter tuning, build/deployment, purely factual queries, general software engineering.
 
 ### Gate 1 · Environment Signal
-The workspace contains architecture-level core code (attention/transformer/MoE, `*.cu`/kernel) or research notes. Routine files like `model.py` or `trainer.py` alone **do not** constitute an environment signal.
+The workspace contains architecture-level core code (attention/transformer/MoE, `*.cu`/kernel) or research notes, **or** cryptography-related code / protocol descriptions / security proof drafts. Routine files like `model.py` or `trainer.py` alone **do not** constitute an environment signal.
 
 ### Gate 2 · Task Signal
-The user's task involves **designing/improving** a new architecture/operator, **analyzing** theoretical properties, **transferring** mathematical structures into AI design, or **querying math knowledge relevant to AI research** (e.g., "how is tangent space used in optimization?"). Pure encyclopedic math queries (e.g., "what is a group?" with no AI context) do not auto-trigger, but can be accessed via `/ask`.
+The user's task involves **designing/improving** a new architecture/operator, **analyzing** theoretical properties, **transferring** mathematical structures into AI design, **analyzing cryptographic constructions/security definitions/reduction proofs/protocols**, or **querying math knowledge relevant to AI research** (e.g., "how is tangent space used in optimization?"). Pure encyclopedic math or cryptography fact queries do not auto-trigger, but can be accessed via `/ask`.
 
 ### Gate 3 · Intent Match
 The user's intent matches one of scenarios A/B/C/D. Pure engineering tasks matching scenario E → no intervention.
 
 > **`/ask` entry**: Manual invocation skips Gate 1 and Gate 2, executing only Gate 0 (exclusion) + Gate 3 (intent match), allowing direct access to any scenario including knowledge queries.
 
+## Domain Router (new in v3.2.0)
+
+> AI research and cryptography **share** mathematical foundations but each has **exclusive** specialty layers. After intent diagnosis and before lens invocation, Domain Router determines the problem's domain and decides which anchors/books/design patterns to load, avoiding cross-domain pollution and token waste.
+
+### Three-Layer Domain Classification
+
+| Layer | Signal Keyword Examples | Loaded Content | Exclusive/Shared |
+|-------|------------------------|----------------|-------------------|
+| **AI Research Layer** | attention, loss, routing, representation, compression, MoE, transformer, KV-cache, LoRA, SSM, diffusion, RL | `../../knowledge-base/` (7 domains, 31 anchors) + `../../design-patterns/` (5 types, 22 patterns) + 7 AI books | AI-exclusive |
+| **Cryptography Layer** | encryption, signature, MAC, PRF/PRG/PRP, OWF, CCA, CPA, AE, zero-knowledge, reduction proof, attack game, DL/CDH/DDH, RSA, ECC, lattice crypto | 3 crypto books + shared math anchors (on demand) + temporary knowledge cards | Crypto-exclusive |
+| **Shared Math Layer** | probability, information theory, entropy, group, ring, field, matrix, spectrum, optimization, convexity, perturbation, complexity | Corresponding anchors in `../../knowledge-base/` + `../../lenses/` lenses | Shared |
+
+### Routing Rules
+
+1. **Judge domain first**: Determine primary domain (AI / Crypto / pure math query) from user keywords.
+2. **No redundant loading of shared math**: If the domain is cryptography, load shared math anchors (e.g., `../../knowledge-base/probability/entropy.md`, `../../knowledge-base/matrix-analysis/spectral-decomposition.md`) on demand; do **NOT** load AI-exclusive `../../design-patterns/`.
+   - **"On demand" criterion**: Load if and only if the problem's mathematical structure maps to that shared anchor's core definition/formulas — i.e., the problem statement explicitly mentions the anchor's core concept (e.g., "spectrum," "entropy," "convex," "perturbation"), or a lens/critic explicitly routes to it. **The domain tag does not decide whether shared anchors load; the problem structure does.**
+3. **Explicit annotation on cross-domain**: If the problem is genuinely AI×crypto intersection (e.g., "use PRF for model watermarking," "reduction proof for adversarial examples"), Domain Router explicitly lists both domains' loaded items and annotates intersection points.
+   - **Intersection annotation template** (4-tuple, feeds critic dim 19 checkpoint 6):
+     1. **Crypto primitive + security property** (e.g., "PRF + pseudorandomness")
+     2. **AI module + functional requirement** (e.g., "watermark + unique traceability")
+     3. **Transfer direction** (crypto→AI / AI→crypto)
+     4. **Assumption achievability after transfer** (Is the original assumption still achievable in the AI scenario? E.g., "Is the PRF assumption satisfiable in ML deployment?")
+4. **No pollution when not cross-domain**: Pure AI problems do not load cryptography books; pure crypto problems do not load AI design patterns. Avoids token waste and conceptual confusion.
+5. **Domain-tagged gap protocol**: Temporary knowledge cards generated by Knowledge Gap Protocol are tagged with domain (AI/Crypto/Shared) for subsequent upgrade to corresponding formal cards.
+
+### Domain Router Decision Flow
+
+```
+User question
+  ↓
+[Gate 0-3 triggered?]
+  ↓ yes
+Domain Router: keyword-based primary domain judgment
+  ├─ AI research → load knowledge-base + design-patterns + AI books
+  ├─ Cryptography → load crypto books + shared math anchors (on demand)
+  ├─ Pure math → only load lenses + corresponding knowledge-base anchors
+  └─ AI×Crypto → dual-domain load + intersection annotation
+  ↓
+[Scenario A/B/C/D routing]
+  ↓
+[Lenses → anchors/books → design translation (AI only) / reduction template (crypto only) → critic]
+```
+
 ## Main Workflow
 
 ### Step 1: Diagnose Intent
 1. Determine which scenario (A/B/C/D/E) the user's intent belongs to
-2. Extract the core tension of the problem: what to preserve? what to suppress? what are the constraints? what is the engineering bottleneck?
-3. Output a problem-type classification
+2. **Domain Router judgment**: problem domain (AI / Crypto / pure math / intersection)
+3. Extract the core tension of the problem: what to preserve? what to suppress? what are the constraints? what is the engineering bottleneck?
+4. Output a problem-type classification + domain tag
 
 ### Step 2: Route Invocation
 
 ```
 Scenario A (Analysis): Select 1–3 lenses → output perspective diagnosis → critic review
 Scenario B (Design): Select 1–3 lenses → invoke relevant activation anchors; if no coverage, enter Knowledge Gap Protocol → generate formal/temporary design patterns → critic review
-Scenario C (Query): Prefer loading relevant activation anchors; if no coverage, generate temporary knowledge card → output per knowledge activation protocol
+  · AI domain: design patterns from design-patterns/; output attention/loss/routing/representation/compression
+  · Crypto domain: design patterns from cryptography books' construction paradigms (SPN/Feistel/Merkle-Damgård/KEM-DEM/Fiat-Shamir); output encryption/MAC/signature/protocols
+Scenario C (Query): Prefer loading relevant activation anchors or crypto books; if no coverage, generate temporary knowledge card → output per knowledge activation protocol
 Scenario D (Verification): Load relevant anchors or temporary knowledge cards → critic reviews conditions and boundaries
+  · AI domain: pass GPU Eight-Dimension Acceptance Gate
+  · Crypto domain: pass reduction tightness + assumption dependency + implementation pitfall checks (GPU gate not required)
 Scenario E (Engineering): No intervention
 ```
 
 ### Step 3: Output Format
 
-**Token-economy rule**: The following is the maximum structure, not the default full template. Trim to the user's question; for simple knowledge queries, provide only the needed definition / formula / risk. Expand design and GPU sections only when relevant, and do not restate loaded cards verbatim.
+**Token-economy rule**: The following is the maximum structure, not the default full template. Trim to the user's question; for simple knowledge queries, provide only the needed definition / formula / risk. Expand design and GPU/reduction sections only when relevant, and do not restate loaded cards verbatim. **After Domain Router determines the domain, only expand the domain-specific subsection.**
 
 **Scenario A/B Output**:
 1. **[Diagnosis]** Problem type + core tension
@@ -137,7 +165,7 @@ See the quantitative checklist in `../../references/gpu-friendly-math.en.md`.
 ## Depth-of-Consultation Protocol
 
 - **Light**: Read knowledge cards (`../../knowledge-base/*/*.en.md`); self-contained and immediately usable
-- **Medium**: Read distilled book notes (`../../references/books/*.en.md`) for more complete context
+- **Medium**: Read distilled book notes (`../../references/books/*.en.md`) for more complete context; for cryptography, also consult the 3 English-language `.md` files listed in `../../references/skill-index.en.md`
 - **Deep**: When `math_book/<PDF>` is available locally, the agent automatically runs `pdftotext` + grep to locate the original page
 
 ## Knowledge Gap Protocol

package/skills/math-research-activator/SKILL.md

@@ -1,8 +1,8 @@
 ---
 name: math-research-activator
 description: |
-  数学研究操作系统：自动诊断用户意图，路由到思想透镜、数学知识库或设计翻译层。触发于设计/改进模型架构/算子/注意力、分析理论性质、迁移数学结构到 AI 设计。不触发于纯工程任务（debug、重构、调参）。
-  English: Mathematical research OS — auto-diagnoses user intent, routes to thinking lenses, math knowledge base, or design translation layer. Triggers on architecture/operator design, theoretical analysis, math-to-AI transfer. Does NOT trigger for pure engineering tasks.
+  数学研究操作系统：自动诊断用户意图，路由到思想透镜、激活锚点或设计翻译层。触发于设计/改进模型架构/算子/注意力、分析理论性质、迁移数学结构到 AI 设计，以及密码学安全定义、构造、归约证明与协议分析。不触发于纯工程任务（debug、重构、调参）。
+  English: Mathematical research OS — auto-diagnoses user intent, routes to thinking lenses, activation anchors, or design translation layer. Triggers on architecture/operator design, theoretical analysis, math-to-AI transfer, and cryptographic definitions, constructions, reductions, or protocol analysis. Does NOT trigger for pure engineering tasks.
 ---
 
 > **语言路由与混合输入规则**：看句式/动词/语气词主框架判定主语言。AI/数学/工程术语不计入。代码/路径/公式不计入。中英接近时沿用上一轮，无上下文默认中文。显式要求优先。中文→本文件，英文→`SKILL.en.md`。完整规则见 `../../references/skill-index.md`。
@@ -11,7 +11,7 @@ description: |
 
 > "思想系统不负责给定理，知识系统不负责乱启发，设计层不负责装深刻。"
 
-本系统是面向 AI 架构创新的数学参谋部——不是武器库，而是告诉你：**这场仗是什么仗、该用什么兵种、怎么部署、哪里会翻车。**
+本系统是面向 AI 架构创新与密码学研究的数学参谋部——不是武器库，而是告诉你：**这场仗是什么仗、该用什么兵种、怎么部署、哪里会翻车。**
 
 ## 核心原则
 
@@ -26,11 +26,11 @@ description: |
 | 层 | 职责 | 目录 | 核心问题 |
 |----|------|------|---------|
 | **思想透镜** | 诊断问题结构，推荐数学视角 | `../../lenses/*.md` | 这个问题该用什么视角看？ |
-| **数学知识** | 提供具体数学工具（定义/定理/公式） | `../../knowledge-base/*/*.md` | 这个视角需要哪些具体数学？ |
+| **激活锚点** | 激活高频数学结构，并在不足时触发 Knowledge Gap Protocol | `../../knowledge-base/*/*.md` | 这个视角需要激活哪些数学结构？ |
 | **设计翻译** | 把数学变成 AI 模块/loss/算子 | `../../design-patterns/*/*.md` | 这些数学怎么变成模型结构？ |
 
 辅助层：
-- `../../references/books/*.md`：7 本书的蒸馏稿，需要深入时的完整上下文
+- `../../references/books/*.md`：10 本书的蒸馏稿，需要深入时的完整上下文；其中 3 本密码学书稿见 `../../references/skill-index.md`
 - `../../references/gpu-friendly-math.md`：GPU 八维验收门（唯一权威）
 - `../../agents/math-critic.md`：数学-工程双重批判器
 
@@ -39,16 +39,16 @@ description: |
 | 场景 | 诊断信号 | 调用路径 |
 |------|---------|---------|
 | **A. 问题分析** | "这个设计合理吗？""逻辑链有没有漏洞？" | 透镜 → critic |
-| **B. 机制设计** | "设计新 attention""把 X 迁移到 Y" | 透镜 → 知识 → 设计 → critic |
-| **C. 知识查询** | "流形上的切空间是什么？""投影定理怎么用？" | 知识 |
-| **D. 验证审查** | "这个公式成立吗？""loss 能保证什么？" | 知识 → 相关设计模式（若引用具体 AI 构造）→ critic |
+| **B. 机制设计** | "设计新 attention""把 X 迁移到 Y" | 透镜 → 激活锚点 → 设计 → critic |
+| **C. 知识查询** | "流形上的切空间是什么？""投影定理怎么用？" | 激活锚点 |
+| **D. 验证审查** | "这个公式成立吗？""loss 能保证什么？" | 激活锚点 → 相关设计模式（若引用具体 AI 构造）→ critic |
 | **E. 纯工程** | debug、重构、调参、代码审查 | **不调用数学系统** |
 
 ## 透镜库（15 个数学视角）
 
 15 个透镜覆盖：公理化、对偶、对称性、谱分解、几何、投影与分解、变分、局部到整体、拓扑、范畴化、扰动、因果、博弈、概率统计、算法。目录：`../../lenses/*.md`。完整目录表见 `../../references/skill-index.md`。
 
-## 知识库（按数学领域组织）
+## 激活锚点（按数学领域组织）
 
 7 个领域：矩阵分析、最优化、微分几何、李理论、拓扑、概率与信息、信息几何。目录：`../../knowledge-base/*/*.md`。完整目录表见 `../../references/skill-index.md`。
 
@@ -64,36 +64,85 @@ description: |
 以下任务**无论工作区含什么**都不触发：代码审查、debug、重构、调参、构建部署、纯事实查询、通用软件工程。
 
 ### Gate 1 · 环境信号
-工作区含架构核心代码（attention/transformer/MoE、`*.cu`/kernel）或研究笔记。仅 `model.py`、`trainer.py` 等常规文件**不构成**环境信号。
+工作区含架构核心代码（attention/transformer/MoE、`*.cu`/kernel）或研究笔记，**或**密码学相关代码/协议描述/安全证明草稿。仅 `model.py`、`trainer.py` 等常规文件**不构成**环境信号。
 
 ### Gate 2 · 任务信号
-用户任务涉及**设计/改进**新架构/算子、**分析**理论性质、**迁移**数学结构到 AI 设计，或**查询与 AI 研究相关的数学知识**（如"切空间在优化中怎么用"）。纯百科式数学查询（如"什么是群"且无 AI 上下文）不自动触发，但可通过 `/ask` 手动进入。
+用户任务涉及**设计/改进**新架构/算子、**分析**理论性质、**迁移**数学结构到 AI 设计、**分析密码学构造/安全定义/归约证明/协议**，或**查询与 AI 研究相关的数学知识**（如"切空间在优化中怎么用"）。纯百科式数学或密码学事实查询不自动触发，但可通过 `/ask` 手动进入。
 
 ### Gate 3 · 意图匹配
 用户意图匹配场景 A/B/C/D 之一。纯工程任务匹配场景 E → 不介入。
 
 > **`/ask` 入口**：手动调用时跳过 Gate 1 和 Gate 2，仅执行 Gate 0（排除门）+ Gate 3（意图匹配），可直接进入任意场景包括知识查询。
 
+## Domain Router（v3.2.0 新增）
+
+> AI 研究与密码学**共享**数学根基，但**独有**各自的专业层。Domain Router 在意图诊断后、调用透镜前，先判定问题归属，决定加载哪些锚点/书稿/设计模式，避免跨域污染与 token 浪费。
+
+### 三层归属判定
+
+| 层 | 信号词示例 | 加载内容 | 独有/共用 |
+|----|-----------|---------|----------|
+| **AI 研究层** | attention、loss、routing、representation、compression、MoE、transformer、KV-cache、LoRA、SSM、扩散、RL | `../../knowledge-base/`（7 领域 31 锚点）+ `../../design-patterns/`（5 类 22 模式）+ AI 方向 7 本书 | AI 独有 |
+| **密码学层** | 加密、签名、MAC、PRF/PRG/PRP、OWF、CCA、CPA、AE、零知识、归约证明、攻击游戏、DL/CDH/DDH、RSA、ECC、格密码 | 3 本密码学书稿 + 共用数学锚点（按需）+ 临时知识卡 | 密码独有 |
+| **共用数学层** | 概率、信息论、熵、群、环、域、矩阵、谱、优化、凸性、扰动、复杂度 | `../../knowledge-base/` 中的对应锚点 + `../../lenses/` 透镜 | 共用 |
+
+### 路由规则
+
+1. **先判 domain**：从用户关键词判定主 domain（AI / 密码 / 纯数学查询）。
+2. **共用数学不重复加载**：若 domain 是密码学，共用数学锚点（如 `../../knowledge-base/probability/entropy.md`、`../../knowledge-base/matrix-analysis/spectral-decomposition.md`）按需加载，**不**加载 AI 专属的 `../../design-patterns/`。
+   - **"按需"判定条件**：当且仅当问题的数学结构映射到该共用锚点的核心定义/公式时加载。即问题陈述中显式出现该锚点对应的核心概念（如"谱""熵""凸""扰动"），或透镜路由/critic 明确指向该锚点。**domain 标签不决定共用锚点加载与否，问题结构决定。**
+3. **跨域时显式标注**：若问题确实是 AI×密码交叉（如"用 PRF 做模型水印""对抗样本的归约证明"），Domain Router 显式列出两个 domain 的加载项，并标注交叉点。
+   - **交叉点标注模板**（四元组，供 critic 第 19 维第 6 检查点审查）：
+     1. **密码学原语 + 安全性质**（如"PRF + 伪随机性"）
+     2. **AI 模块 + 功能需求**（如"水印 + 唯一可追踪性"）
+     3. **迁移方向**（密码→AI / AI→密码）
+     4. **迁移后假设可达性**（原假设在 AI 场景是否仍可达成？如"PRF 的 PRF 假设在 ML 部署中是否可满足"）
+4. **不跨域时不污染**：纯 AI 问题不加载密码学书稿；纯密码学问题不加载 AI 设计模式。避免 token 浪费与概念混淆。
+5. **缺口协议分 domain**：Knowledge Gap Protocol 生成的临时知识卡标注 domain（AI/密码/共用），便于后续升级到对应正式卡片。
+
+### Domain Router 判定流程图
+
+```
+用户问题
+  ↓
+[Gate 0-3 触发?]
+  ↓ 是
+Domain Router: 关键词判定主 domain
+  ├─ AI 研究 → 加载 knowledge-base + design-patterns + AI 书稿
+  ├─ 密码学  → 加载密码学书稿 + 共用数学锚点（按需）
+  ├─ 纯数学  → 只加载 lenses + 对应 knowledge-base 锚点
+  └─ AI×密码 → 双 domain 加载 + 交叉点标注
+  ↓
+[场景 A/B/C/D 路由]
+  ↓
+[透镜 → 锚点/书稿 → 设计翻译（仅 AI）/ 归约模板（仅密码）→ critic]
+```
+
 ## 主流程
 
 ### 第一步：诊断意图
 1. 判断用户意图属于场景 A/B/C/D/E 哪个
-2. 提取问题核心张力：想保留什么？想抑制什么？约束是什么？工程瓶颈是什么？
-3. 输出问题类型分类
+2. **Domain Router 判定**：问题归属（AI / 密码 / 纯数学 / 交叉）
+3. 提取问题核心张力：想保留什么？想抑制什么？约束是什么？工程瓶颈是什么？
+4. 输出问题类型分类 + domain 标注
 
 ### 第二步：路由调用
 
 ```
 场景 A（分析）：选 1-3 个透镜 → 输出视角诊断 → critic 审查
 场景 B（设计）：选 1-3 个透镜 → 调用相关激活锚点；若无覆盖则进入 Knowledge Gap Protocol → 生成正式/临时设计模式 → critic 审查
-场景 C（查询）：优先加载相关激活锚点；若无覆盖则生成临时知识卡 → 按知识激活协议输出
+  · AI domain：设计模式来自 design-patterns/，产出 attention/loss/routing/representation/compression
+  · 密码 domain：设计模式来自密码学书稿的构造范式（SPN/Feistel/Merkle-Damgård/KEM-DEM/Fiat-Shamir），产出加密/MAC/签名/协议
+场景 C（查询）：优先加载相关激活锚点或密码学书稿；若无覆盖则生成临时知识卡 → 按知识激活协议输出
 场景 D（验证）：加载相关锚点或临时知识卡 → critic 审查条件与边界
+  · AI domain：过 GPU 八维验收门
+  · 密码 domain：过归约紧度 + 假设依赖 + 实现陷阱检查（不必过 GPU 门）
 场景 E（工程）：不介入
 ```
 
 ### 第三步：输出格式
 
-**Token 经济原则**：以下是最长结构，不是默认全文模板。按用户问题裁剪；简单知识查询只给必要定义/公式/风险，设计与 GPU 内容只在与问题有关时展开；避免复述已加载卡片全文。
+**Token 经济原则**：以下是最长结构，不是默认全文模板。按用户问题裁剪；简单知识查询只给必要定义/公式/风险，设计与 GPU/归约内容只在与问题有关时展开；避免复述已加载卡片全文；**Domain Router 已判定 domain 后，只展开该 domain 的专属小节**。
 
 **场景 A/B 输出**：
 1. **[诊断]** 问题类型 + 核心张力
@@ -137,7 +186,7 @@ description: |
 ## 深度查阅协议
 
 - **轻度**：读知识卡片（`../../knowledge-base/*/*.md`），自足可用
-- **中度**：读书蒸馏稿（`../../references/books/*.md`），获取更完整上下文
+- **中度**：读书蒸馏稿（`../../references/books/*.md`），获取更完整上下文；密码学问题优先查阅 `../../references/skill-index.md` 列出的 3 本专门书稿
 - **深度**：本机有 `math_book/<PDF>` 时，Agent 自动 `pdftotext` + grep 定位原文页
 
 ## 知识缺口协议 / Knowledge Gap Protocol